Performance Evaluation of Elliptical Journal Bearing Using ANN with Micropolar Lubricants

This analysis presents an artificial neural network (ANN)-based framework for predicting the steady-state performance characteristics of journal bearings lubricated with micropolar fluids, with validation against conventional numerical methods. Initially, the Reynolds equation governing micropolar lubrication is solved using the Finite Difference Method (FDM). The caviatation (i.e. Swift–Stieber) boundary condition is applied to obtain pressure distributions, enabling accurate evaluation of steady-state performance characteristics under varying operating conditions, including eccentricity ratio (ε), characteristic length of micropolar fluid (l_m), and coupling number (N²).

The numerical results generated through FDM are subsequently utilized to develop ANN models based on a dataset comprising more than 1560 input–output pairs. Three feed-forward ANN architectures—Levenberg–Marquardt, Bayesian regularization, and Scaled Conjugate Gradient—are implemented, and their predictive capabilities are systematically evaluated through comparison with FDM results.

The findings demonstrate that ANN models achieve substantial reductions in computational time and effort while maintaining high prediction accuracy. The maximum deviation between ANN predictions and FDM results is within 5%, indicating strong agreement and reliability. Furthermore, the trained ANN models exhibit robust generalization capability across the considered parameter range. Overall, the proposed ANN framework offers an efficient and accurate alternative for performance prediction and facilitates rapid analysis and design optimization of micropolar-lubricated journal bearing systems.